Jerry Breecher

7: Deadlocks1

OPERATING SYSTEMS

DEADLOCKS

What Is In This Chapter? What is a deadlock?

Staying Safe: Preventing and Avoiding Deadlocks

Living Dangerously: Let the deadlock happen, then detect it and recover from it.

7: Deadlocks2

OPERATING SYSTEM Deadlocks

 EXAMPLES:  "It takes money to make money".

You can't get a job without experience; you can't get experience without a job.

 BACKGROUND:

The cause of deadlocks: Each process needing what another process has. This results from sharing resources such as memory, devices, links.

Under normal operation, a resource allocations proceed like this:: 

1. Request a resource (suspend until available if necessary ).2. Use the resource.3. Release the resource.

7: Deadlocks3

Traffic only in one direction. Each section of a bridge can be viewed as a resource. If a deadlock occurs, it can be resolved if one car backs up

(preempt resources and rollback). Several cars may have to be backed up if a deadlock occurs. Starvation is possible.

7: Deadlocks4

DEADLOCKS Bridge Crossing Example

NECESSARY CONDITIONSALL of these four must happen simultaneously for a deadlock to occur:

7: Deadlocks5

DEADLOCK CHARACTERISATION

Mutual exclusionOne or more than one resource must be held by a process in a non-sharable (exclusive) mode. Hold and WaitA process holds a resource while waiting for another resource. No PreemptionThere is only voluntary release of a resource - nobody else can make a process give up a resource. Circular WaitProcess A waits for Process B waits for Process C .... waits for Process A.

A visual ( mathematical ) way to determine if a deadlock has, or may occur. 

G = ( V, E ) The graph contains nodes and edges. V Nodes consist of processes = { P1, P2, P3, ...} and

resource types { R1, R2, ...}

 E Edges are ( Pi, Rj ) or ( Ri, Pj )

 An arrow from the process to resource indicates the process is requesting the resource. An arrow from resource to process shows an instance of the resource has been allocated to the process. Process is a circle, resource type is square; dots represent number of instances of resource in type. Request points to square, assignment comes from dot.

7: Deadlocks6

RESOURCE ALLOCATION GRAPH

PiRj

PiRj

Pi

If the graph contains no cycles, then no process is deadlocked. If there is a cycle, then:

a) If resource types have multiple instances, then deadlock MAY exist.b) If each resource type has 1 instance, then deadlock has occurred.

7: Deadlocks7

RESOURCE ALLOCATION GRAPH

Resource allocation graph

P2 Requests P3

R3 Assigned to P3

7: Deadlocks8

RESOURCE ALLOCATION GRAPH

Resource allocation graphwith a deadlock.

Resource allocation graphwith a cycle but no deadlock.

HOW TO HANDLE DEADLOCKS – GENERAL STRATEGIES There are three methods: Ignore Deadlocks:

Ensure deadlock never occurs using either 

Prevention Prevent any one of the 4 conditions from happening.

  Avoidance Allow all deadlock conditions, but calculate cycles

about to happen and stop dangerous operations..  Allow deadlock to happen. This requires using both: 

Detection Know a deadlock has occurred.  Recovery Regain the resources.

7: Deadlocks9

DEADLOCKS Strategy

Most Operating systems do this!!

 Do not allow one of the four conditions to occur.

Mutual exclusion:a) Automatically holds for printers and other non-sharables.b) Shared entities (read only files) don't need mutual exclusion (and

aren’t susceptible to deadlock.)c) Prevention not possible, since some devices are intrinsically non-

sharable. Hold and wait:

a) Collect all resources before execution.b) A particular resource can only be requested when no others are

being held. A sequence of resources is always collected beginning with the same one.

c) Utilization is low, starvation possible. 

7: Deadlocks10

DEADLOCKS DeadlockPrevention

 Do not allow one of the four conditions to occur. No preemption: 

a) Release any resource already being held if the process can't get an additional resource.

b) Allow preemption - if a needed resource is held by another process, which is also waiting on some resource, steal it. Otherwise wait.

 Circular wait: 

a) Number resources and only request in ascending order.b) EACH of these prevention techniques may cause a decrease in

utilization and/or resources. For this reason, prevention isn't necessarily the best technique.

c) Prevention is generally the easiest to implement.

7: Deadlocks11

DEADLOCKS DeadlockPrevention

If we have prior knowledge of how resources will be requested, it's possible to determine if we are entering an "unsafe" state. Possible states are: 

Deadlock No forward progress can be made. Unsafe state A state that may allow deadlock. Safe state A state is safe if a sequence of processes exist such that

there are enough resources for the first to finish, and as each finishes and releases its resources there are enough for the next to finish.

 The rule is simple: If a request allocation would cause an unsafe state, do not honor that request. NOTE: All deadlocks are unsafe, but all unsafes are NOT deadlocks.

7: Deadlocks12

DEADLOCKS Deadlock Avoidance

NOTE: All deadlocks are unsafe, but all unsafes are NOT deadlocks.

7: Deadlocks13

SAFEDEADLOCK

UNSAFE

Only with luck will processes avoid

deadlock.

O.S. can avoid deadlock.

DEADLOCKS Deadlock Avoidance

Let's assume a very simple model: each process declares its maximum needs. In this case, algorithms exist that will ensure that no unsafe state is reached. EXAMPLE:There exists a total of 12 tape drives. The current state looks like this:

7: Deadlocks14

In this example, < p1, p0, p2 > is a workable sequence.

Suppose p2 requests and is given one more tape drive. What happens then?

Process Max Needs Allocated Current Needs

P0 10 5 5

P1 4 2 2

P2 9 2 7

DEADLOCKS DeadlockAvoidance

There are multiple instances of the resource in these examples.

A method used to determine if a particular state is safe. It's safe if there exists a sequence of processes such that for all the processes, there’s a way to avoid deadlock:

The algorithm uses these variables: 

Need[I] – the remaining resource needs of each process.Work - Temporary variable – how many of the resource are currently

available.Finish[I] – flag for each process showing we’ve analyzed that process

or not.

need <= available + allocated[0] + .. + allocated[I-1] <- Sign of success

 Let work and finish be vectors of length m and n respectively. 

7: Deadlocks15

DEADLOCKS

Safety Algorithm

DeadlockAvoidance

1. Initialize work = availableInitialize finish[i] = false, for i = 1,2,3,..n

 2. Find an i such that:

finish[i] == false and need[i] <= work 

If no such i exists, go to step 4. 3. work = work + allocation[i]

finish[i] = truegoto step 2

 4. if finish[i] == true for all i, then the system is in a safe state.

7: Deadlocks16

DEADLOCKS DeadlockAvoidance

Safety Algorithm

Do these examples:Consider a system with: five processes, P0 P4, three resource types, A, B, C.Type A has 10 instances, B has 5 instances, C has 7 instances. At time T0 the following snapshot of the system is taken. 

7: Deadlocks17

Is the system in a safe state?

DEADLOCKS DeadlockAvoidanceSafety Algorithm

134200P4

110112P3

006203P2

020002P1

233347010P0

CBACBACBA

AvailReqAlloc

Max Needs = allocated + can-be-requested

Do these examples:Now try it again with only a slight change in the request by P1. P1 requests one additional resource of type A, and two more of type C. Request1 = (1,0,2).Is Request1 < available? 

7: Deadlocks18

Alloc Req Avail

A B C A B C A B C

P0 0 1 0 7 4 3 1# 3 0#

P1 3# 0 2# 0 2 0

P2 3 0 2 6 0 0

P3 2 1 1 0 1 1

P4 0 0 2 4 3 1

Produce the state chart as if the

request is Granted and see if it’s safe.(We’ve drawn the

chart as if it’s granted.

DEADLOCKS DeadlockAvoidance

Safety Algorithm

Can the request be granted?

Need an algorithm that determines if deadlock occurred.

 Also need a means of recovering

from that deadlock.

7: Deadlocks19

DEADLOCKS Deadlock Detection

SINGLE INSTANCE OF A RESOURCE TYPE • Wait-for graph == remove the resources

from the usual graph and collapse edges.• An edge from p(j) to p(i) implies that p(j) is

waiting for p(i) to release. 

SEVERAL INSTANCES OF A RESOURCE TYPE Complexity is of order m * n * n. We need to keep track of: 

available - records how many resources of each type are available.allocation - number of resources of type m allocated to process n.request - number of resources of type m requested by process n.

 Let work and finish be vectors of length m and n respectively. 

7: Deadlocks20

DEADLOCKS Deadlock Detection

1. Initialize work[] = available[] For i = 1,2,...n, if allocation[i] != 0 then finish[i] = false; otherwise, finish[i] = true; 2. Find an i such that: finish[i] == false and request[i] <= work  If no such i exists, go to step 4. 3. work = work + allocation[i] finish[i] = true goto step 2 4. if finish[i] == false for some i, then the system is in deadlock state. IF finish[i] == false, then process p[i] is deadlocked.

7: Deadlocks21

DEADLOCKS Deadlock Detection

EXAMPLEWe have three resources, A, B, and C. A has 7 instances, B has 2 instances, and C has 6 instances. At this time, the allocation, etc. looks like this: 

7: Deadlocks22

Is there a sequence that will allow deadlock to be avoided?

Is there more than one sequence that will work?

200200P4

001112P3

000303P2

202002P1

000000010P0

CBACBACBA

AvailReqAlloc

DEADLOCKS Deadlock Detection

EXAMPLE Suppose the Request matrix is changed like this. In other words, the maximum amounts to be allocated are initially declared so that this request matrix results. 

7: Deadlocks23

USAGE OF THIS DETECTION ALGORITHM Frequency of check depends on how often a deadlock occurs and how many processes will be affected.

Is there now a sequence that will allow deadlock to be avoided?

200200P4

001112P3

1#00303P2

202002P1

000000010P0

CBACBACBA

AvailReqAlloc

DEADLOCKS Deadlock Detection

So, the deadlock has occurred. Now, how do we get the resources back and gain forward progress?

 PROCESS TERMINATION: 

Could delete all the processes in the deadlock -- this is expensive. Delete one at a time until deadlock is broken ( time consuming ). Select who to terminate based on priority, time executed, time to

completion, needs for completion, or depth of rollback In general, it's easier to preempt the resource, than to terminate the

process. RESOURCE PREEMPTION: 

Select a victim - which process and which resource to preempt. Rollback to previously defined "safe" state. Prevent one process from always being the one preempted ( starvation ).

7: Deadlocks24

DEADLOCKS Deadlock Recovery

COMBINED APPROACH TO DEADLOCK HANDLING:  Type of resource may dictate best deadlock handling. Look at ease of

implementation, and effect on performance.

In other words, there is no one best technique.

Cases include: 

Preemption for memory, Preallocation for swap space, Avoidance for devices ( can extract Needs from process. )

 

7: Deadlocks25

DEADLOCKS Deadlock Recovery

In this section we have:

Looked at necessary conditions for a deadlock to occur.

Determined how to prevent, avoid, detect and recover from deadlocks.

7: Deadlocks26

WRAPUP